The Multimodal Shift: Why Health AI Needs More Than Hospital Data
Author: Peehu Sachdeva, Market Strategy Lead, Axia Medicine
Published by Axia Medicine | November 2025
This is the second article of a series, following "The Genomics Paradox".
In the last decade, the UK Biobank showed what a large, high-quality multimodal dataset can do for research, producing thousands of publications. It also showed how slow and expensive that progress can be. It took a decade for UK Biobank to become the UK Biobank. Few efforts have come close: FinnGen, All of Us, Our Future Health and a handful of others.
In the private sector, Tempus AI’s IPO created a global multi-omics asset, albeit limited to oncology and requiring billions in investment. If every therapeutic area required comparable efforts to make multimodal data accessible, the industry would spend a century waiting for the datasets needed today.
The impact of current initiatives is undeniable; equally undeniable is that we cannot afford to keep waiting for the next breakthrough. And as we explored in The Genomic Paradox, this lag in discovery is not technical, it is structural. If slogans like "health data is commoditized" were true, how hard could it be to make diverse, multimodal data widely available?
Today's data collection systems are simply unable to fix the shortage of multimodal and multiomics data: they were built for institutions, not for individuals.
And in Precision Medicine, data is always singular.
From Short Reads to Multimodal Intelligence
The Genomics Paradox highlighted that sequencing is abundant but access remains scarce. Each new sensor, wearable, and assay generates insight into human biology, yet most measurements end up in silos behind institutional or commercial walls.
No slogan can bridge these gaps. Even the most advanced model cannot learn from data it cannot see, and synthetic data still depends on real patient data to exist. To reach its potential, AI must be native to multimodality, integrating genomics, proteomics, imaging, clinical notes and behavioural data from inception, not as an add-on.
When GSK, Regeneron, and UK Biobank combined 500,000 exomes with longitudinal health records, the result unlocked discoveries unreachable by either modality alone. That partnership proved that the availability of multimodal data can drive progress across entire industries, globally.
The Data We Ignore in Everyday Health
Consider a young diabetes patient. Their record captures pristine billing codes and lab results but says nothing about the food desert they live in or the stress of working two jobs. Roughly 3 in 4 U.S. adults now live with one or more chronic conditions (Figure 1), many of which are influenced by nutrition, genetics and environment. Yet the most advanced health systems still operate on fragments of information.
Even beyond omics, most of our health data lives outside the clinic: in behaviour, in our daily routines and environmental exposure. That is the inevitable truth: care delivery still relies on a fraction of the patient’s story. Industry keeps optimising for what is easy to measure, missing what truly determines outcomes.
With such a narrow foundation, even the most sophisticated health-AI model is forced to guess, building predictions for billions of dynamic patients from a few million static snapshots. When Health AI learns from partial truths, hallucinations are only the tip of the iceberg.
Figure 1. U.S. trends in single and multiple chronic conditions (2013-2023). Most age groups show rising prevalence over time. Solid lines mark meaningful changes; dashed lines indicate no significant trend.
Source: Centers for Disease Control and Prevention. Trends in Multiple Chronic Conditions Among US Adults, By Life Stage, Behavioral Risk Factor Surveillance System, 2013–2023.
Native Multimodal AI and Data Liquidity
Most platforms today are retrofitted for multiple data types, losing the relationships that make multimodal data valuable. Native architectures preserve those connections in real time, allowing AI to detect how gene variants, diet, air quality, and social context interact to shape health outcomes.
But broader access is unlikely to come from isolated institutional silos alone. It will require patient-driven infrastructure, as we will highlight in our fourth article: networks where individuals can contribute and control their data directly, with consent and transparency built in.
Patients already generate the data needed for this shift through diagnostics, wearables, and research. The challenge is not collection, but flow: ensuring data can move safely to the studies that need it. That is true data liquidity: information that moves without losing context or trust.
It demands three design principles: Ownership stays with the individual. Traceability makes every use auditable. Shared value ensures benefits for both patients and science.
Once patients are visible as individuals rather than tokens, engagement becomes continuous. AI can learn what outreach works, refine participation, and build diversity over time. That is how multimodal AI compounds in value and reliability for the long term, starting with individual patients and, as Precision Public Health details, encompassing the wider population.
Participatory Medicine: Health AI's First Job
The promise of multimodal AI is not smarter prediction, but smarter Participation. When biological, behavioural and environmental data intersect, we can understand not only who gets sick but why, where, and when. And when the infrastructure is in place to unlock the full picture of a patient's health, only when Health AI will be able to deliver on its promises.
Because in Healthcare, AI's first job is to see each patient as an individual: this is what Precision Medicine has always been about, from Day One.
References
Ardic, N. and Dinc, R. (2025). Emerging trends in multi-modal artificial intelligence for clinical decision support: A narrative review. PubMed, 31(3), p.14604582251366141-14604582251366141. doi:https://doi.org/10.1177/14604582251366141.
Bijan Mossadeghi, Caixeta, R., Ondarsuhu, D., Luciani, S., Hambleton, I.R. and Hennis, A.J.M. (2023). Multimorbidity and social determinants of health in the US prior to the COVID-19 pandemic and implications for health outcomes: a cross-sectional analysis based on NHANES 2017–2018. BMC Public Health, 23(1). doi:https://doi.org/10.1186/s12889-023-15768-8.
Chustecki, M. (2024). Benefits and Risks of AI in Health Care: Narrative Review. Interactive Journal of Medical Research, [online] 13(e53616). doi:https://doi.org/10.2196/53616.
El Arab, R.A., Abu-Mahfouz, M.S., Abuadas, F.H., Alzghoul, H., Almari, M., Ghannam, A. and Seweid, M.M. (2025). Bridging the Gap: From AI Success in Clinical Trials to Real-World Healthcare Implementation—A Narrative Review. Healthcare, 13(7), p.701. doi:https://doi.org/10.3390/healthcare13070701.
Hao, Y., Cheng, C., Li, J., Li, H., Di, X., Zeng, X., Jin, S., Han, X., Liu, C., Wang, Q., Luo, B., Zeng, X. and Li, K. (2025). Multimodal Integration in Healthcare: Development with Applications in Disease Management (Preprint). Journal of Medical Internet Research. [online] doi:https://doi.org/10.2196/76557.
Poon, H. (2023). Multimodal Generative AI for Precision Health. NJEM AI. [online] doi:https://ai.nejm.org/doi/10.1056/AI-S2300233.
Schouten, D., Nicoletti, G., Dille, B., Chia, C., Vendittelli, P., Schuurmans, M., Litjens, G. and Khalili, N. (2025). Navigating the landscape of multimodal AI in medicine: A scoping review on technical challenges and clinical applications. Medical image analysis, [online] 105, p.103621. doi:https://doi.org/10.1016/j.media.2025.103621.
Shaik, T., Tao, X., Li, L., Xie, H. and Velásquez, J.D. (2024). A survey of multimodal information fusion for smart healthcare: Mapping the journey from data to wisdom. Information Fusion, [online] 102, p.102040. doi:https://doi.org/10.1016/j.inffus.2023.102040.
Simon, B., Ozyoruk, K., Gelikman, D., Harmon, S. and Türkbey, B. (2024). The future of multimodal artificial intelligence models for integrating imaging and clinical metadata: A narrative review. Diagn Interv Radiol. [online] doi:https://doi.org/10.4274/dir.2024.242631.
Sudlow, C., Gallacher, J., Allen, N., Beral, V., Burton, P., Danesh, J., Downey, P., Elliott, P., Green, J., Landray, M., Liu, B., Matthews, P., Ong, G., Pell, J., Silman, A., Young, A., Sprosen, T., Peakman, T. and Collins, R. (2015). UK Biobank: An Open Access Resource for Identifying the Causes of a Wide Range of Complex Diseases of Middle and Old Age. PLOS Medicine, 12(3), p.e1001779. doi:https://doi.org/10.1371/journal.pmed.1001779.
Van Hout, C.V., Tachmazidou, I., Backman, J.D., Hoffman, J.D., Liu, D., Pandey, A.K., Gonzaga-Jauregui, C., Khalid, S., Ye, B., Banerjee, N., Li, A.H., O’Dushlaine, C., Marcketta, A., Staples, J., Schurmann, C., Hawes, A., Maxwell, E., Barnard, L., Lopez, A. and Penn, J. (2020). Exome sequencing and characterization of 49,960 individuals in the UK Biobank. Nature, [online] 586(7831), pp.749–756. doi:https://doi.org/10.1038/s41586-020-2853-0.
Watson, K.B., Wiltz, J.L., Nhim, K., Kaufmann, R.B., Thomas, C.W. and Greenlund, K.J. (2025). Trends in Multiple Chronic Conditions Among US Adults, By Life Stage, Behavioral Risk Factor Surveillance System, 2013–2023. Preventing Chronic Disease, [online] 22.
